International journal Nature has recently published an environmental study where researchers have developed an artificial intelligence (AI) model that shows the potential to improve the accuracy of flood forecasting. The study indicates that the AI system performs comparably or even better than current best practices and could provide earlier warnings of extreme flood events.
According to the paper, human-induced climate change has resulted in increased occurrences of flooding in certain regions, and current forecasting methods rely on river gauges (monitoring stations built along rivers). However, the uneven global distribution of these gauges limits this approach, creating greater challenges in forecasting for ungauged rivers, particularly in developing nations.
Lead author and corresponding author Grey Nearing of Google Research in the United States, along with colleagues and collaborators, trained an AI model on 5680 existing gauges to predict daily river discharge in ungauged basins over a 7-day forecast horizon. This AI model was then benchmarked against the Global Flood Awareness System (GloFAS), the current best practice global software for forecasting short- to medium-range flood hazards. Results show that the AI model can provide flood forecasts with up to five days' lead time and comparable or better accuracy than the existing system on the same day.
Furthermore, the newly developed AI model exhibits comparable or better forecast skill for extreme events with a 5-year return period than GloFAS for events with a 1-year return period.
The study authors conclude that their AI model demonstrates the potential for providing longer lead times for small to extreme floods in ungauged basins, offering the prospect of reliable flood forecasting in developing regions.
